OpenAI Glossary Cheat Sheet

🤖 Artificial General Intelligence (AGI): AGI refers to a hypothetical 🌌 GPT-4: GPT-4 is a hypothetical, more advanced version of the
AI that can perform any intellectual task a human being can do, GPT series, expected to have larger model size and enhanced
demonstrating human-like cognitive abilities across diverse domains. capabilities compared to its predecessors.
🚀 Singularity: A theoretical point in the future when AI 🏋️ Pre-training: Pre-training is the initial phase of training a deep
advancements lead to rapid, uncontrollable, and transformative learning model on a large dataset, often unsupervised
changes in society, potentially surpassing human comprehension.
🎛️ Fine-tuning: Fine-tuning is the process of adapting a pre-
🛡️ AI Safety: AI safety is the study and practice of building AI trained model for a specific task by training it on labeled data
systems that operate securely and align with human values, ensuring related to that task, refining its performance.
that they benefit humanity without causing harm.
🎯 Zero-shot learning: Zero-shot learning is a machine learning
🧭 Alignment Problem: The alignment problem is the challenge of approach where a model can make predictions or complete tasks
designing AI systems that understand and act upon human intentions, without being explicitly trained on that task's data.
values, and goals, rather than optimizing for unintended objectives.
🧪 Few-shot learning: Few-shot learning is a machine learning
🧠 OpenAI: OpenAI is an AI research organization that focuses on approach where a model can quickly adapt to new tasks by
developing artificial general intelligence (AGI) that benefits everybody. learning from a small number of labeled examples.
💡 Deep Learning: Deep learning is a subfield of machine learning 📜 Token: A token is a unit of text, such as a word or subword, that
that uses artificial neural networks to model complex patterns and serves as input to a language model.
make predictions or decisions based on input data.
🔪 Tokenizer: A tokenizer is a tool that breaks down text into
🕸️ Artificial Neural Network: An artificial neural network is a individual tokens for processing by a language model.
computational model inspired by the human brain's structure and 🖼️ Context window: The context window is the maximum number
function, consisting of interconnected nodes called neurons that of tokens that a language model can process in a single pass,
process and transmit information. determining its ability to capture context in input data.
🎓 Supervised Learning: Supervised learning is a machine learning
💡 Prompts: Prompts are input text given to a language model to
approach where a model is trained on a dataset containing input- generate a response or complete a specific task.
output pairs, learning to predict outputs based on new inputs.
🎨 Prompt Engineering: Prompt engineering is the process of
🌐 Unsupervised Learning: Unsupervised learning is a machine designing effective prompts to elicit desired responses from
learning approach where a model learns patterns and structures within language models, improving their utility and reliability.
input data without explicit output labels, often through clustering or
dimensionality reduction. 🤖 ChatGPT: ChatGPT is a conversational AI model developed by
OpenAI based on the GPT architecture, designed to generate
🎮 Reinforcement Learning from Human Feedback (RLHF): RLHF human-like responses in text-based conversations.
is a method that combines reinforcement learning with human
feedback, allowing AI models to learn from and adapt to human 📚 InstructGPT: InstructGPT is an AI model developed by
preferences and values. OpenAI, designed to follow instructions given in prompts, enabling
it to generate more task-specific and accurate responses.
💬 Natural Language Processing (NLP): NLP is a field of AI that
focuses on enabling computers to understand, interpret, and generate 🔧 OpenAI API: The OpenAI API is a service provided by OpenAI
human language. that allows developers to access and utilize their AI models, such
as ChatGPT, for various applications.
📚 Large Language Models: Large language models are AI models
trained on vast amounts of text data, capable of understanding and 🎨 DALL-E: DALL-E is an AI model developed by OpenAI that
generating human-like text. generates images from textual descriptions, combining natural
language understanding with image generation capabilities.
⚙️ Transformer: The Transformer is a deep learning architecture
designed for sequence-to-sequence tasks, known for its self-attention 🐍 LaMDA: LaMDA is Google's conversational AI model designed
mechanism that helps capture long-range dependencies in data. to engage in open-domain conversations, understanding and
generating responses for a wide range of topics.
👁️ Attention mechanism: Attention mechanisms in neural networks
enable models to weigh the importance of different input elements 🧭 Midjourney: Midjourney refers to an intermediate point in the
relative to one another, improving their ability to capture context. development of AI technologies, signifying progress made while
acknowledging that there is still much to be achieved.
🔄 Self-attention: Self-attention is a type of attention mechanism
used in transformers that allows the model to relate different positions 🌊 Stable diffusion: Stable diffusion is a research area focused on
of a single sequence. improving the training of large-scale AI models by introducing
stability and controllability during the diffusion process.
📖 BERT (Bidirectional Encoder Representations from
Transformers): BERT is a pre-trained transformer-based model 📈 Diffusion models: Diffusion models are a class of models that
developed by Google for natural language understanding tasks, which represent the spread of information, influence, or other phenomena
can be fine-tuned for specific applications. through a network.
🚀 GPT (Generative Pre-trained Transformer): GPT is a series of AI 🔄 Backpropagation: Backpropagation is a widely-used
models developed by OpenAI, designed for natural language optimization algorithm in neural networks that minimizes the error
processing tasks and capable of generating coherent, contextually between predicted outputs and true outputs by adjusting the
relevant text. model's weights.
🌐 GPT-3.5: GPT-3.5 is an intermediate version of the GPT series,
bridging the gap between GPT-3 and GPT-4 in terms of model size
and capabilities.

Cheat Sheet Download Link: https://blog.finxter.com/openai-glossary/